Freepik Launches F Lite: An Open AI Image Generator Trained on Licensed Data

Freepik Unveils F Lite: A New Approach to AI Image Generation

Freepik, the popular online graphic design platform, has entered the AI image generation arena with the launch of F Lite, an "open" AI image model that sets itself apart by being trained exclusively on commercially licensed, "safe-for-work" images 1

1

. Developed in collaboration with AI startup Fal.ai, F Lite represents a significant step towards addressing the ongoing copyright concerns in the AI industry.

Technical Specifications and Development

F Lite boasts impressive technical specifications, containing approximately 10 billion parameters. The model's development was a substantial undertaking, utilizing 64 Nvidia H100 GPUs over a two-month training period 1

1

. This process involved an internal dataset of around 80 million images, which, while smaller than some competitors, potentially makes F Lite "the largest publicly available text-to-image model trained entirely on legally sound content" 2

2

.

Two Flavors of F Lite

Freepik has released F Lite in two distinct versions:

1. Standard Version: More predictable and faithful to prompts.
2. Texture Version: More chaotic and error-prone, but capable of delivering better textures and creative compositions 1

   1

   3

   3

   .

Both versions are openly licensed, with weights available on Hugging Face, and the model's code is open-source, allowing for integration into various workflows and customization 2

2

.

Positioning in the AI Image Generation Market

While Freepik doesn't claim superiority over leading image generators like Midjourney's V7 or Black Forest Labs' Flux family, the company's goal was to create a solid, open model that developers can adapt and improve 2

2

. Running F Lite requires substantial computational power, needing a GPU with at least 24GB of VRAM 1

1

.

Addressing Copyright Concerns

F Lite joins a small but growing collection of generative AI models trained on licensed data, including those from Adobe, Bria, Getty Images, Moonvalley, and Shutterstock 1

1

2

2

. This approach stands in contrast to many AI companies that have faced copyright lawsuits for using protected content without permission, often arguing fair use – a stance challenged by many creators and IP rights holders 1

1

.

Potential Impact and Future Outlook

The introduction of F Lite could have significant implications for the AI image generation landscape. As legal battles around AI and copyright continue, demand for models like F Lite – trained entirely on licensed content – may see rapid growth 2

2

. This shift could potentially reshape the industry, especially if corporate entities become wary of the legal uncertainties surrounding more established image generators.

Open-Source Approach

Freepik's decision to make F Lite open-source is noteworthy. The model's code is available for use in ComfyUI, integration into Python workflows via diffusers, and for fine-tuning and creating custom LoRAs 2

2

. This open approach could foster innovation and improvements from the developer community, potentially accelerating the development of ethically trained AI models.

As the AI image generation field continues to evolve, F Lite represents a significant step towards addressing ethical and legal concerns while providing a powerful tool for creators and developers alike.